No announcement yet.

3850 ha/failover

  • Filter
  • Time
  • Show
Clear All
new posts

  • 3850 ha/failover

    I'm not sure if I'm doing something wrong, or misunderstanding capabilities.

    I have two x 3850. I configured one by itself as a core switch, got it all working the way I wanted it.

    I connected the second switch with the stackwise cable, watched it join the stack. All good. I can see if I "lose power" to a switch, then the second one takes over as active.

    however, I think my misunderstanding comes in here.

    I wanted to have a "downstream" switch connected from GI1 to GI1/0/1.
    Then have the same downstream switch connected from GI2 to GI/0/1
    both configured exactly the same. so if the active switch fails, the passive should bring everything up

    actually.. in the process of writing, I think I know what I did wrong.

    I need to configure the Gi1 and Gi2 on the downstream switch in a port-group.
    then configure GI1/0/1 and GI2/0/1 to be the other end of the port-group.

    Correct ? failure of either GI1/01 or 2/0/1 would mean the other interface should start working.. ?
    Please do show your appreciation to those who assist you by leaving Rep Point

  • #2
    Re: 3850 ha/failover

    If you connect the downstream switch to two ports on the upstream switch and neither switch is running Spanning Tree (unlikely), you'll have created a loop and the network will go down as soon as the first broadcast packet appears.

    If the switches are running the original Spanning Tree protocol, you will get redundancy but no load balancing. Since a port on a switch running Spanning Tree has to go through the "listening" and "learning" phase before becoming active, it will take the downstream switch 2 x "forward delay" (standard is 15 seconds) to activate a previously inactive port. 30 seconds is usually way too long for transparent layer 2 failover, but you could reduce the "forward delay" parameter to a more manageable value.

    If the switches are running Rapid Spanning Tree, the delay will be 6 seconds or less. You'll still only get redundancy and no load balancing.

    If you configure the ports as port groups at both ends (LACP or Cisco EtherChannel), you'll get transparent, immediate failover as well as load balancing when both links are active.


    • #3
      Re: 3850 ha/failover

      Agree with Ser Olmy: use port-channel groups for links between the same 2 devices (stacks or single boxes.) Both links are active all the time to double your bandwidth, and act as a failover if one goes down. When running links with stacks to single devices, best to use 1 cable from the single device to each member of the stack. All such links are grouped using LACP at each end, so that Spanning Tree doesn't get in the way, giving you the best redundancy if any stack member fails. Obviously the single device is still a single point of failure, but you've got to draw the line somewhere.

      I use multiple 1Gig links between a core stack and a local access stack in the same building, getting a >1Gig link between without spending for 10Gig tech.
      MSCA (2003/XP), Security+, CCNA

      ** Remember: credit where credit is due, and reputation points as appropriate **


      • #4
        Re: 3850 ha/failover

        thanks guys, just talking it out helped, but appreciate your validation

        i was talking to them about it this morning and in the end, due to a number of reasons (low rack space being one of them, but also the risk of increased complexity by adding multiple cables that allllllll have to be correct means they would prefer a "cold-spare" situation.

        now i just need to work out how to remove the Stacking information. despite removing the stackwise cable and running "no switch 2 provision global" it seems to still think the 2nd member is just inactive..

        ah, no, there we go.
        Last edited by tehcamel; 22nd August 2014, 01:07.
        Please do show your appreciation to those who assist you by leaving Rep Point